Flagged for Scrutiny: A Pilot Randomized Vignette Experiment on Algorithmic Advice and Administrative Burdens

Administrative burdens impede access to social safety net programs and disproportionately affect marginalized groups, reinforcing existing inequalities. Government agencies increasingly deploy algorithmic decision-support tools to streamline processes and improve uptake. Yet growing evidence suggests that these tools can produce disparate impact and may automate inequality. What remains underexplored is how frontline staff interact with algorithmic outputs and whether cognitive biases in human–AI interaction translate into unequal, burden-creating choices. Gules-Guctas will conduct a pilot randomized factorial vignette experiment with staff in a state unemployment insurance agency to test whether algorithmic recommendations causally increase the likelihood that frontline staff impose burden-creating decisions and whether these effects are disproportionately larger for otherwise identical applicants who signal Black or Latino identity.